[LLVMdev] [llvm-c]

[Christian Budde]

Am 14.05.2013 21:42, schrieb Christian Budde:
> Am 13.05.2013 16:16, schrieb Tom Stellard:
>> On Sat, May 11, 2013 at 01:48:44PM +0200, Christian Budde wrote:
>>> Hello,
>>>
>>> a few days ago, I was trying to access the LLVM target registry via the
>>> LLVM-C interface. Unfortunately I can't seem to get some useful
>>> information out of it.
>>>
>>> Despite the fact that the following code is written in Pascal, can you
>>> please tell me if I have done something wrong here?
>>>
>>> // initialize all targets / target information
>>> LLVMInitializeAllTargets;
>>> LLVMInitializeAllTargetInfos;
>>>
>>> // initialize native target in particular
>>> LLVMInitializeNativeTarget;
>>>
>>> // get first target
>>> Target := LLVMGetFirstTarget;
>>>
>>> // loop until the target is NULL
>>> while (Target <> nil) do
>>> begin
>>>   // get target name & description
>>>   TargetName := LLVMGetTargetName(Target);
>>>   TargetDescription := LLVMGetTargetDescription(Target);
>>>
>>>   // do something with these information
>>>
>>>   Target := LLVMGetNextTarget(Target);
>>> end;
>>>
>> I don't see any obvious problems here, though someone else might.
>> What exactly is the problem you are seeing?  Is LLVMGetFirstTarget
>> returning NULL?
> LLVMGetFirstTarget returns a pointer at a quite high address. Subsequent
> calls to LLVMGetNextTarget all return the very same pointer (much
> lower). If I query LLVMGetTargetName and LLVMGetTargetDescription I got
> an empty string and a null pointer.
>
>> You may want to take a look at this code:
>> http://cgit.freedesktop.org/mesa/mesa/tree/src/gallium/drivers/radeon/radeon_llvm_emit.c
>> It is a pretty good example of using the C API for accessing the target
>> registry.  However, if you want to test this code directly, I would
>> recommend replacing the r600 target triple with your native
>> architecture, because the R600 backend is not built by default.
> I've done it exactly like in this example (except the fact that I'm
> using pascal as language and x86 as main target). So if the sequence of
> calls seems to be right, I'll be investigating whether I've done
> something wrong compiling the DLL.

I just found the bug. It was all my fault. When migrating the pascal
interface code  from 3.2 to 3.3, I accidentally removed the explicit
declaration of the calling convention (should have been cdecl).

Now everything works as expected.

Thanks for the clarification,

    Christian

-- 
Christian-W. Budde
Eleonorenstr. 17
30449 Hannover
Tel.: +49 511 31048857
E-Mail: Christian at savioursofsoul.de
WWW: http://www.savioursofsoul.de/Christian